nested or dummy coded?

[Zdaniuk, Bozena-3]

This may be a very naive question but i am hoping people will be kind :)
        I have a few interventionists coming to classes and delivering an intervention to students. Each interventionist is doing multiple classes. To control for and examine potential interventionist impact should I a) employ a 3-level model - students nested within class, classes nested within interventionist? Or should I dummy-code the interventionists and use the dummy vars as additional predictors at the "student" level?
Bozena

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Maguin, Eugene]

Bozena,
I went back and looked at the other messages you posted on what I assume is the same project and it seems that you have a pre-post treatment-control design with random assignment at the class level (20 classes total) and multiple classes being taught/led by the same person.

I know Ryan said that you should consult with an expert. You might find the experience level you need on the Multilevel listserv.

I'm not an expert. I want to bring one point that you may have already thought of and decided about to your attention, which is that having multiple classes taught by the same person creates a 'mixed' design in that the intervenor grouping level applies to the treatment arm but not to the control arm. This has been written about in Psych Methods (I don't have the citation here nor do I recall a useful search phrase). Setting that issue aside, my understanding of recommended sample sizes for level 2 analyses is that 20 is at the lower level. Introducing an intervenor level reduces that number, almost certainly by a factor of two and maybe more. My impression is this project is on the design stage and not the analysis stage. With data you could compute ICCs and use those results to inform your analyses. Thus it seems that treating intervenor as a covariate is a better plan. That said, it seems to me that intervenor is a level 2 variable because everybody in a class will have t!
 he same intervenor code; thus, no variance.

Gene Maguin


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Zdaniuk, Bozena
Sent: Tuesday, October 01, 2013 4:58 PM
To: [hidden email]
Subject: nested or dummy coded?

This may be a very naive question but i am hoping people will be kind :)
        I have a few interventionists coming to classes and delivering an intervention to students. Each interventionist is doing multiple classes. To control for and examine potential interventionist impact should I a) employ a 3-level model - students nested within class, classes nested within interventionist? Or should I dummy-code the interventionists and use the dummy vars as additional predictors at the "student" level?
Bozena

=====================
To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD